I’ve been using Gemini 3.5 Pro in the Google AI IDE and it keeps doing the same thing no matter what I try. It refuses to write complete code. It’s less frequent with Gemini 3 Flash, but it still happens.
I’ve been explicit about it. No placeholders, no TODOs, no mocks, no fake helpers. I’ve said production code, real scenario, real codebase. I’ve rewritten prompts, tightened them, simplified them, spelled everything out. It makes no difference.
The output is always half-finished. Comments where logic should be. “You can implement this part yourself” energy. It reads like a Stack Overflow answer that stops right before it becomes useful. That’s not what I expect from an AI that’s embedded directly into an IDE and marketed at developers.
This isn’t an occasional miss, it feels like the default behavior. At this point it doesn’t seem like a prompting issue at all. It feels like the model is actively avoiding committing to full solutions.
Claude Sonnet/Opus is the only model I’ve used that will actually reason through a problem and give end-to-end, usable code without constantly backing out. Google has started rate limiting it pretty hard, so if this placeholder issue doesn’t get fixed I’m going straight to Claude code.
I’m posting to see if anyone else is running into this, or if this is just how Gemini is expected to behave inside the IDE right now.